A simple analytical criterion is provided for deciding whether a human or automation is best for a failure detection task. The method is based on expected-value decision theory in much the same way as is signal detection. It requires specification of the probabilities of misses (false negatives) and false alarms (false positives) for both human and automation being considered, as well as factors independent of the choice -- namely, costs and benefits of incorrect and correct decisions as well as the prior probability of failure. The method can also serve as a basis for comparing different modes of automation. Some limiting cases of application are discussed, as are some decision criteria other than expected value. Actual or potential applications include the design and evaluation of any system in which either humans or automation are being considered.
Whatever the task, however large or small, the problem of deciding when to have a human operator perform a task and when to automate has frustrated the engineering community for decades. One approach (that of the technology protagonist) is to automate whenever technically possible. Another (perhaps that of the humanist) is to automate when the task is boring, physically risky, or otherwise unpleasant and undesirable for a human but to retain for the human
such work as is satisfying. A third (an economist's perspective) is to automate when automation is cheaper than human labor, by whatever economic measure one chooses. A fourth perspective is that no general criteria or rational method can be provided, that allocation of human or automation is and perhaps forever will be an art (Sheridan, 1998a). A fifth perspective asserts that the decision of whether or not to automate should depend on circumstances of the moment and should be altered dynamically on the basis of factors such as human operator workload and s ituation awareness (Hancock & Scallen, 1996; for more detailed discussions and comparisons of these various perspectives, see Billings, 1996; Hancock & Scallen, 1996; Parasuraman & Riley, 1997; Sheridan, 1992).
This brief paper rejects the first approach, except when demonstration of technological prowess has intrinsic value, as it might for some scientific or entertainment reasons. Although the second approach is compelling, boredom, risk, pleasure, and satisfaction can all, in principle, be put in either monetary or subjectively expected utility equivalents. We accept the premises of the fourth and fifth perspectives in spirit but claim that there exists some rationality and stability in the allocation of a task to human versus automation, provided the task is sufficiently simple and well defined. So we choose the third approach, that of economics.
How can an economic approach be applied to the question of whether or not to automate a task? Expected-value analysis provides a Standard method, but having a simple and well-defined task is key to applying the method. Evaluation of performance in most tasks involves considerations of benefits and costs that accrue as a function of whether the response is appropriate or inappropriate under many different circumstances, the occurrences of which lie outside the control of the operator - whether human or automation. For any given task there certainly are many different circumstances that arise, and for each there are many degrees of response appropriateness.
To make the problem tractable, we selected a failure-response task that can be simplified to only two circumstances, each of which is associated with true but uncontrollable states of the world - namely, failure and normal - and two responses, one appropriate to failure and one appropriate to normal. This results in a two-by-two set of circumstances. The fuse blows, the pipe breaks, the check bounces, or it does not. The response is appropriate, or it is not. For each of the four possibilities there are probabilities, benefits, and costs, and the response can be automated or not. …